The Lancet Oncology Manuscript Draft Manuscript Number: THELANCETONCOLOGY-D-13-00486 Title: Air pollution and lung cancer incidence in 17 European cohorts within the ESCAPE study Article Type: Articles (Original Research) Keywords: air pollution, particulate matter, lung cancer, adenocarcinoma Abstract: Background: Ambient air pollution is suspected to cause lung cancer. We studied associations between long-term residential exposure to air pollution and lung cancer incidence in European populations. Methods: We used individual data for 17 European cohorts. Baseline addresses were geocoded and air pollution was assessed by land-use regression models for particulate matter (PM) below 10 µm (PM10), below 2.5 µm (PM2.5), between 2.5 and 10 µm (PMcoarse), PM2.5absorbance, nitrogen oxides and two traffic indicators. We used Cox regression models with adjustment for potential confounders for cohort-specific analyses and random effects models for meta-analyses. Findings: The 312 944 cohort members contributed 4 013 131 person-years at risk. During follow-up (average 12.8 years), 2095 incident lung cancer cases were diagnosed. The meta-analyses showed that a 10-µg/m3 increase in PM10 was associated with a 22% (95% confidence interval [CI]: 3-45%) greater risk for lung cancer, and a 5-µg/m3 increase in PM2.5 was associated with an 18% (95% CI: -4 to 46%) greater risk. The same increments of PM10 and PM2.5 were associated with 49% (95% CI: 8-105%) and 55% (95% CI: 5-129%) higher risk for adenocarcinomas of the lung, respectively. An increase in road traffic of 4000 vehicle-km/day within 100 m of the residence was associated with a 9% (95% CI: -1 to 21%) higher risk for lung cancer. Associations between PM air pollution metrics and lung cancer were also detected among never smokers and below current European Union limit values for PM. The results showed no association with nitrogen oxides or traffic intensity on the nearest street. Interpretation: Particulate matter air pollution contributes to lung cancer incidence in Europe. Funding: The European Community's Seventh Framework Program (FP7/2007-2011) under grant agreement number: 211250.

1

Air pollution and lung cancer incidence in 17 European cohorts within the

ESCAPE study

Ole Raaschou-Nielsen PhD1*

, Zorana J Andersen PhD 1,2

, Rob Beelen PhD3, Evangelia

Samoli PhD4, Massimo Stafoggia MSc

5, Gudrun Weinmayr PhD

6,7, Barbara Hoffmann MD

7,8,

Paul Fischer MSc9, Mark J Nieuwenhuijsen PhD

10, Bert Brunekreef PhD

3,11, Wei W Xun

MPH12

, Klea Katsouyanni PhD4, Konstantina Dimakopoulou MSc

4, Johan Sommar MSc

13,

Bertil Forsberg PhD13

, Lars Modig PhD13

, Anna Oudin PhD13

, Bente Oftedal PhD14

, Per E

Schwarze PhD14

, Per Nafstad MD14,15

, Ulf De Faire PhD16

, Nancy L Pedersen PhD17

, Claes-

Göran Östenson PhD18

, Laura Fratiglioni PhD19

, Johanna Penell PhD16

, Michal Korek MSc16

,

Göran Pershagen PhD16

, Kirsten T Eriksen PhD1, Mette Sørensen PhD

1, Anne Tjønneland

DMSc1, Thomas Ellermann PhD

20, Marloes Eeftens MSc

3, Petra H Peeters PhD

11, Kees

Meliefste BSc3, Meng Wang MSc

3, Bas Bueno-de-Mesquita PhD

21, Timothy J Key DPhil

22,

Kees de Hoogh PhD23

, Hans Concin MD24

, Gabriele Nagel PhD6,24

, Alice Vilier MSc25,26,27

,

Sara Grioni BSc28

, Vittorio Krogh MD28

, Ming-Yi Tsai PhD29,30

, Fulvio Ricceri PhD31

,

Carlotta Sacerdote PhD32

, Claudia Galassi MD32

, Enrica Migliore MSc32

, Andrea Ranzi

PhD33

, Giulia Cesaroni MSc5, Chiara Badaloni MSc

5, Francesco Forastiere PhD

5, Ibon

Tamayo MSc34

, Pilar Amiano MSc35

, Miren Dorronsoro MD35

, Antonia Trichopoulou MD4,36

,

Christina Bamia PhD4, Paolo Vineis MPH

12 †, Gerard Hoek PhD

3†

1Danish Cancer Society Research Center, Strandboulevarden 49, 2100 Copenhagen, Denmark

2Center for Epidemiology and Screening, Department of Public Health, University of

Copenhagen, Øster Farimagsgade 5, 1014 Copenhagen K, Denmark

Manuscript

2

3Institute for Risk Assessment Sciences, Utrecht University, PO Box 80178, 3508 TD

Utrecht, The Netherlands

4Department of Hygiene, Epidemiology and Medical Statistics, Medical School, National and

Kapodistrian University of Athens, Mikras Asias 75, 11527 Athens, Greece

5Department of Epidemiology, Lazio Regional Health Service, Local Health Unit ASL RME,

Via S.Costanza 53, 00198 Rome, Italy

6Institute of Epidemiology and Medical Biometry, Ulm University, Helmholtzstr. 22, 89081

Ulm, Germany

7IUF – Leibniz Research Institute for Environmental Medicine, Auf’m Hennekamp 50, 40225

Düsseldorf, Germany

8Medical Faculty, Heinrich Heine University of Düsseldorf, D-40225 Düsseldorf, Germany

9National Institute for Public Health and the Environment, Center for Sustainability and

Environmental Health, P.O. Box 1, 3720 BA Bilthoven, The Netherlands

10Center for Research in Environmental Epidemiology, Parc de Recerca Biomèdica de

Barcelona – PRBB (office 183.05), C. Doctor Aiguader, 88, 08003 Barcelona, Spain

11Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht,

Universiteitsveg 100, 3584 CG Utrecht, The Netherlands

12MRC-HPA Centre for Environment and Health, Department of Epidemiology and

Biostatistics, Imperial College London, St Mary’s Campus, Norfolk Place W2 1PG, London,

United Kingdom

13Division of Occupational and Environmental Medicine, Department of Public Health and

Clinical Medicine, Umeå University, SE-90187 Umeå, Sweden

14Norwegian Institute of Public Health, 4404 Nydalen, Oslo 0403, Norway

15Institute of Health and Society, University of Oslo, Pb 1130 Blindern 0318 Oslo, Norway

16Institute of Environmental Medicine, Karolinska Institute, 17177 Stockholm, Sweden

3

17Department of Medical Epidemiology and Biostatistics, Karolinska Institute, 17177

Stockholm, Sweden

18Department of Molecular Medicine and Surgery, Karolinska Institutet, Karolinska

University Hospital, SE-17176 Stockholm, Sweden

19Aging Research Center, Department of Neurobiology, Care Sciences and Society,

Karolinska Institute and Stockholm University, Gävlegatan 16, 11330 Stockholm, Sweden

20Department of Environmental Science, Aarhus University, Frederiksborgvej 399, 4000

Roskilde, Denmark

21National Institute for Public Health and the Environment, Antonie van Leeuwenhoeklaan 9,

Bilthoven, P.O. Box 1, 3720 BA Bilthoven, The Netherlands

22Cancer Epidemiology Unit, Nuffield Department of Clinical Medicine, University of

Oxford, Richard Doll Building, Roosevelt Drive, Oxford OX3 7LF, United Kingdom

23MRC-HPA Centre for Environment and Health, Department of Epidemiology and

Biostatistics, Imperial College London, St Mary’s Campus, Norfolk Place, London W2 1PG

24Agency for Preventive and Social Medicine, Rheinstrase 61, 6900 Bregenz, Austria

25Inserm, Centre for Research in Epidemiology and Population Health, U 1018, Nutrition,

Hormones and Women’s Health team, F-94805 Villejuif, France

26University Paris Sud, UMRS 1018, F-94805, Villejuif, France

27IGR, F-94805, Villejuif, France

28 Epidemiology and Prevention Unit, Fondazione IRCCS Istituto Nazionale dei Tumori –

Milan, Via Venezian 1, 20133 Milan, Italy

29Department of Epidemiology and Public Health, Swiss Tropical & Public Health Institute,

Basel, Switzerland, University of Basel, Basel, Switzerland

30Department of Environmental & Occupational Health Sciences, University of Washington,

Box 357234 Seattle, USA

4

31Human Genetics Foundation, I-10126 – Via Nizza 52, Turin, Italy

32Unit of Cancer Epidemiology, AO Citta’ della Salute e della Scienza-University of Turin

and Center for Cancer Prevention, Via Santena 7, 10126 Turin, Italy

33Environmental Health Reference Centre – Regional Agency for Environmental Prevention

of Emilia-Romagna, Via Begarelli 13, 41121 Modena, Italy

34 Health Division of Gipuzkoa, Research Institute of BioDonostia, Avenida de Navarra 4,

Donostia-San Sebastian, Spain

35CIBERESP, Consortium for Biomedical Research in Epidemiology and Public Health,

Madrid, Spain

36Hellenic Health Foundation, Kaisareias 13 & Alexandroupoleos GR-115 27, Athens, Greece

* Corresponding author at Danish Cancer Society Research Center, Strandboulevarden 49,

2100 Copenhagen Ø; email: ole@cancer.dk; telephone: +45 35257617

† Shared last-authorship

5

SUMMARY

Background: Ambient air pollution is suspected to cause lung cancer. We studied associations

between long-term residential exposure to air pollution and lung cancer incidence in European

populations.

Methods: We used individual data for 17 European cohorts. Baseline addresses were

geocoded and air pollution was assessed by land-use regression models for particulate matter

(PM) below 10 µm (PM10), below 2·5 µm (PM2.5), between 2·5 and 10 µm (PMcoarse),

PM2.5absorbance, nitrogen oxides and two traffic indicators. We used Cox regression models

with adjustment for potential confounders for cohort-specific analyses and random effects

models for meta-analyses.

Findings: The 312 944 cohort members contributed 4 013 131 person-years at risk. During

follow-up (average 12·8 years), 2095 incident lung cancer cases were diagnosed. The meta-

analyses showed that a 10-µg/m3 increase in PM10 was associated with a 22% (95%

confidence interval [CI]: 3–45%) greater risk for lung cancer, and a 5-µg/m3 increase in PM2.5

was associated with an 18% (95% CI: –4 to 46%) greater risk. The same increments of PM10

and PM2.5 were associated with 49% (95% CI: 8–105%) and 55% (95% CI: 5–129%) higher

risk for adenocarcinomas of the lung, respectively. An increase in road traffic of 4000

vehicle-km/day within 100 m of the residence was associated with a 9% (95% CI: –1 to 21%)

higher risk for lung cancer. Associations between PM air pollution metrics and lung cancer

were also detected among never smokers and below current European Union limit values for

PM. The results showed no association with nitrogen oxides or traffic intensity on the nearest

street.

6

Interpretation: Particulate matter air pollution contributes to lung cancer incidence in Europe.

Funding: The European Community’s Seventh Framework Program (FP7/2007-2011) under

grant agreement number: 211250.

7

INTRODUCTION

Lung cancer is one of the most frequent cancers with a dismal prognosis. Active smoking is

the principal cause but occupational exposures, residential radon, environmental tobacco

smoke and lower socio-economic status are also established risk factors. Ambient air

pollution, specifically particulate matter (PM) with absorbed polycyclic aromatic

hydrocarbons and other genotoxic chemicals, is suspected to increase the risk for lung cancer.

Several epidemiological studies have shown higher risks for lung cancer in association with

various measures of air pollution1-10

and indicated an association mainly in non-3,11

and never-

12,13 smokers and in individuals with low fruit consumption.

3,12 In the western world, overall

lung cancer incidence rates have stabilized during the past decades, but major shifts have been

recorded in the frequencies of different histological types of lung cancer, with considerable

relative increases in adenocarcinomas and decreases in squamous cell carcinomas.14

Changes

in tobacco blends14

and ambient air pollution15,16

might have influenced these shifts.

Within the ESCAPE project, we analysed data from 17 European cohort studies with a wide

range of exposure levels in order to investigate the hypotheses that 1) residential air pollution

(specifically particulate matter) is associated with the risk for lung cancer, 2) the association

between air pollution and risk for lung cancer is stronger among non-smokers and people with

low fruit intake, and 3) the association with air pollution is stronger for adenocarcinomas and

squamous cell carcinomas than for all lung cancers combined.

8

METHODS

The association between long-term exposure to air pollution and incidence of lung cancer was

analysed in each cohort separately at the local centre by common standardized protocols for

exposure assessment, outcome definition, confounder models and statistical analyses. Cohort-

specific effect estimates were combined by meta-analysis at the Danish Cancer Society

Research Center, Copenhagen, Denmark. A pooled analysis of all cohort data was not

possible due to data-transfer and privacy issues.

Cohorts

The 17 cohorts were in Sweden (EPIC-Umeå, SNACK-K, SALT, Sixty, SDPP), Norway

(HUBRO), Denmark (DCH), the Netherlands (EPIC-MORGEN, EPIC-PROSPECT), the

United Kingdom (EPIC-Oxford), Austria (VHM&PP), Italy (EPIC-Varese, EPIC-Turin,

SIDRIA-Turin, SIDRIA-Rome), Spain (EPIC-San Sebastian) and Greece (EPIC-Athens)

(Figure 1). More details of each cohort are given in an online appendix. The study areas were

mostly large cities with suburban or rural communities. Some of the cohorts covered large

regions of the country, such as EPIC-MORGEN in the Netherlands, EPIC-Oxford in the

United Kingdom and the VHM&PP cohort in Austria. For DCH, EPIC-Oxford, VHM&PP

and EPIC-Athens, exposure could be assigned to part of the original cohort, and only those

parts were analysed.

Definition of incident lung cancer cases

The main outcome was all cancers of the lung; secondary analyses addressed

adenocarcinomas and squamous cell carcinomas of the lung. We included cancers located in

9

the bronchus and the lung (ICD10/ICDO3: C34.0–C34.9). We only included primary cancers

(i.e. not metastases). Each cancer was histologically characterized and squamous cell

carcinomas (ICDO3: 8050–8084 and 5th morphology code digit = 3) and adenocarcinomas

(ICDO3: 8140–8384 and 5th morphology code digit = 3) were singled out. Lymphomas in the

lung (ICDO3 morphology codes 9590/3–9729/3) were not included. The cohort members

were followed up for cancer incidence in national or local cancer registries, except for EPIC-

Athens, where cases were identified by questionnaires and telephone interviews followed by

verification of medical records, and the SIDRIA cohorts, for which hospital discharge and

mortality register data were used.

Exposure assessment

Air pollution concentrations at the baseline residential addresses of study participants were

estimated by land use regression models in a three-step, standardized procedure. First, PM

with an aerodynamic diameter < 10 µm (PM10), PM with aerodynamic diameter < 2·5 µm

(PM2.5), PM2.5absorbance (a marker for black carbon and soot), nitrogen oxides (NOx), and

nitrogen dioxide (NO2) were measured during different seasons at locations for each cohort

population between October 2008 and April 2011.17,18

Coarse PM was calculated as the

difference between PM10 and PM2.5. In three areas only NOx was measured. Second, land-use

regression models were developed for each pollutant in each study area, with the annual mean

concentration as the dependent variable and an extensive list of geographical attributes as

possible predictors.19,20

The models generally explained a large fraction of measured spatial

variation, the R2 from leave-one-out cross-validation usually falling between 0·60 and 0·80

(Table S1, online appendix). Finally, the models were used to assess exposure at the baseline

address of each cohort member. We also collected information on two indicators of traffic at

10

the residence: traffic intensity (vehicles/day) on the nearest street and total traffic load

(vehicle-km driven per day) on all major roads within 100 m.

Statistical analyses

Proportional hazards Cox regression models were fitted for each cohort, with age as the

underlying time scale. Participants were followed up for lung cancer from enrolment until the

time of a lung cancer diagnosis or censoring. Participants with a cancer (except non-

melanoma skin cancer) before enrolment were excluded. Censoring was done at the time of

death, a diagnosis of any other cancer (except non-melanoma skin cancer), emigration,

disappearance, loss to follow-up for other reasons or end of follow-up, whichever came first.

For the analyses of histological subtypes of lung cancer, cases of different histological

subtypes were censored.

Air pollution exposure was analysed as a linear variable in three a-priori specified confounder

models. Model 1 included gender, calendar time and age (time axis). Model 2 additionally

adjusted for smoking status (never/former/current), smoking intensity, (smoking intensity)2,

smoking duration, time since quitting smoking, environmental tobacco smoke, occupation,

fruit intake, marital status, educational level and employment status. Model 3 (the main

model) further adjusted for area-level socio-economic status. A cohort was included only if

information on age, gender, calendar time, smoking status, smoking intensity and smoking

duration were available. Table S2 (online appendix) shows the available variables for each

cohort.

We evaluated individual characteristics as a-priori potential effect modifiers: age (< 65, ≥ 65),

gender, educational level, smoking status, fruit intake (<150, 150–300, ≥300 g/day). Age was

11

analysed time-dependently. For a few cohorts (HUBRO, Sixty, SDPP) for which there was

information about fruit intake in categories such as 'a few times per week', 'daily', 'several

times per day', the lowest category was analysed with < 150 g/day, the medium category with

150–300 g/day and the highest category with ≥ 300 g/day.

A number of sensitivity analyses and model checks were conducted for each cohort, all with

confounder model 3. First, we restricted the analyses to participants who had lived at the

baseline address throughout the follow-up period in order to minimize misclassification of

long-term exposure relevant to the development of lung cancer. Secondly, we added an

indicator of degree of urbanization to model 3. Thirdly, we tested the linear assumption in the

relation between each air pollutant and lung cancer by replacing the linear term with a natural

cubic spline with three equally spaced inner knots, and compared the model fit of the linear

and the spline models by the likelihood-ratio test. Fourthly, to investigate if an association

between air pollution and risk for lung cancer was detectable below a-priori defined

thresholds, models were run including only participants exposed to air pollution

concentrations below those thresholds.

In the meta-analysis, we used random-effects models to pool the results for cohorts.21

I2

statistics 22

and p-values for the 2 test from Cochran’s Q were calculated to investigate the

heterogeneity among cohort-specific effect estimates. Effect modification was tested by meta-

analysing the pooled estimates from the different strata with the 2 test of heterogeneity. We

evaluated the robustness of the results by repeating the meta-analysis after exclusion of the

two largest cohorts.

We used a common STATA (www.stata.com) script for all analyses, except for spline

models, which were fitted with R software (www.r-project.org).

12

Role of the funding source

The funding source had no role in the study. The authors had full access to all data and

decided independently to submit the paper.

13

RESULTS

Seventeen cohorts in nine European countries contributed to this study. Altogether 312 944

cohort members contributed 4 013 131 person-years at risk and 2095 incident lung cancer

cases that developed during follow-up (average, 12·8 years). The number of participants and

cases varied considerably, the Danish and Austrian cohorts contributing more than half the

cases (Table 1). The cohort areas represented a wide range of exposures, with 3–12 times

higher mean air pollution concentrations in some southern than in some northern areas (Table

1). The variation in exposure within study areas was substantial (Figure 2 and Figure S1,

online appendix). The mean age at enrolment varied from 43 to 73 years (Table 1).

The meta-analysis showed associations that were statistically significant or of borderline

significance between PM10 (hazard ratio (HR, 1·22 per 10 µg/m3; 95% confidence interval

(CI): 1·03-1·45), PM2.5 (HR, 1·18 per 5 µg/m3; 95% CI: 0·96-1·46) and traffic load at major

roads within 100 m (HR, 1·09 per 4000 vehicle-km/day; 95% CI: 0·99-1·21) and the risk for

lung cancer in confounder model 3 (Table 2). The results from model 1, with adjustment only

for age, sex, and calendar time, showed stronger associations; the effect of adjustment was

due mainly to the smoking variables. No association was found with NO2, NOx or traffic

intensity at the nearest street. Restriction to the 14 cohorts for whom estimates of exposure to

PM were available gave similar results for NO2 (HR, 1·01; 95% CI: 0·94-1·09) and NOx (HR,

1·03; 95% CI: 0·97-1·10). Figure 3 shows the HRs for each cohort and from the meta-analyses

for PM10 and PM2.5. Although the HRs varied across cohorts, the 95% CIs for each cohort

always included the overall meta-analysis estimate, and there was no significant heterogeneity

between cohorts. Figure S2 (online appendix) shows plots for the other air pollutants and the

traffic indicators. Table 3 shows stronger, statistically significant associations between PM10

and PM2.5 and adenocarcinomas of the lung than for all cancers. Restriction to participants

14

who had lived at the same residence throughout the follow-up period gave consistently

stronger associations both for all lung cancers and for adenocarcinomas (Table 3). The

stronger associations with adenocarcinomas and for people who had not moved were not due

to selection of cohorts contributing to these results (Table 3). No significant association was

found with squamous cell carcinomas.

Table 4 shows that restriction of participants to those with exposure below any of the pre-

defined thresholds for particulate matter concentrations provided fairly stable elevated HRs.

This finding is complemented by the results of the spline models (Table S3, online appendix),

showing that the association between air pollution and risk for lung cancer did not deviate

statistically significantly from linear.

Table S4 (online appendix) shows no clear differences between the HRs for lung cancer

associated with PM10 and PM2.5 by gender, age, educational level, smoking status or fruit

intake, with widely overlapping CIs for the effect modifier levels; all the p-values for

interaction were ≥ 0·19. Elevated HRs for lung cancer in association with PM10 and PM2.5

were also observed among never-smokers.

Table S5 (online appendix) shows that the HR for lung cancer in association with PM10 and

PM2.5 was virtually identical after exclusion of the two largest cohorts, which contributed the

majority of the cases. Adjustment for degree of urbanization led to a small change in the HR

for PM10, which was, however, due almost entirely to selection of contributing cohorts and

not to adjustment for urbanization per se.

15

Discussion

This analysis of 17 European cohort studies shows associations between residential exposure

to PM air pollution at enrolment and the risk for lung cancer. The associations were stronger

for adenocarcinomas of the lung and among participants who lived at the enrolment address

throughout the follow-up period.

The strengths of our study include the use of 17 cohort studies in multiple locations in Europe

with very different exposure levels and also the use of standardized protocols for exposure

assessment and data analysis. A comprehensive set of pollutants was assessed, in contrast to

many previous studies; few previous European studies assessed PM air pollution. Individual

exposure assessment was based on actual measurements made in the development of land-use

regression models for the detection of within-area contrasts. The study benefits from the

standardized exposure assessment, a large number of participants, and information on

potential confounders.

Most previous cohort studies of ambient PM air pollution and lung cancer incidence or

mortality in general populations showed associations that were statistically significant or of

borderline significance,1,4-8,10,23,24

whereas two studies found no such association.12,25

The

present study, one of the largest of its kind, estimated a 40% (95% CI: –8 to 113%) increase

in risk per 10 µg/m3 PM2.5, which is similar to the Six-City estimate in the USA of a 37%

(95% CI: 7–75%) increase,7 but higher than the estimate from the American Cancer Society

study (14%; 95% CI: 4–23%),1 and from studies in the Netherlands (–19%; 95% CI: –37 to

4%),12

Japan (24%; 95% CI: 12–37%),4 China (3%; 95% CI: 0–7%),

5 and Italy (5%; 95% CI:

1–10%).10

The confidence intervals of these estimates, however, overlap with ours, so that the

differences could be due to random variation. Previously estimated associations with PM10

differ more widely than those with PM2.5. Our estimated 22% (95% CI: 3–45%) increase in

16

risk per 10 µg/m3 PM10 is in line with that of a recent study in New Zealand (15%; 95% CI:

4–26%),6 higher than that in a previous European study (–9%; 95% CI: –30 to 18%)

25 and

lower that those in studies in the USA (421%; 95% CI: 94–1299%) per 24 µg/m3 PM10),

23 and

Germany (84%; 95% CI: 23–174%) per 7 µg/m3 PM10).

8 In most of the previous studies,

exposure was monitored at a central site; few estimated exposure at individual addresses, as

was done in our study.

In the western world, relative increases in incidence rates of adenocarcinomas and decreases

for squamous cell carcinomas of the lung has occurred in recent decades.14

Changes in

cigarette design may have influenced these shifts: filtered cigarettes and changes in tobacco

blends have decreased the exposure of smokers to polycyclic aromatic hydrocarbons and tar

and increased their exposure to nitrates and toxic agents formed from nitrogen oxides.14

Studies of time trends and geographical correlations have suggested that ambient air pollution

might also have influenced the incidence of adenocarcinomas,15,16

whereas one study

suggested an association between air pollution and squamous cell carcinomas of the lung.13

The present study confirms a stronger association between air pollution and adenocarcinomas.

Our study has some limitations. It is difficult to disentangle the effects of single air pollutants

in an epidemiological study because they are part of complex mixtures; however, it seems

likely that PM is the most important component for cancer risk. In agreement with this notion,

diesel engine exhaust was recently classified as a human carcinogen by the International

Agency for Research on Cancer.26

PM in ambient air, with absorbed polycyclic aromatic

hydrocarbons, transition metals and other substances, is capable of causing oxidative stress,

inflammation, and direct and indirect genotoxicity.27,28

Associations with PM rather than with

nitrogen oxides thus appear to be plausible.

17

We used land use regression models to estimate exposure at the baseline address; however,

even the best exposure models incorporate some degree of misclassification. Any

misclassification is expected to be non-differential and consequently to bias the estimated

HRs towards the null. We used data on air pollution for 2008–2011 in developing our land

use regression models but applied them to baseline addresses mainly 10–15 years earlier.

Recent work in Rome, the Netherlands and Vancouver has shown that the spatial distribution

of air pollution is relatively stable over 10–year periods.29

In our study, exposure was assessed

at the enrolment address; moving from that address during follow-up might lead to

misclassification of the exposure relevant to later development of lung cancer. Our results

show stronger associations between air pollution and the risk for lung cancer among people

who lived at the same address throughout follow-up. The latency for lung cancer can be

several decades;30

our results indicate that more recent exposure is also important.

The cohort-specific analyses consistently identified smoking-related variables as the most

important confounders, in accordance with the fact that smoking is the most important risk

factor for lung cancer. Information on smoking variables was available for all the cohorts, and

we would expect only weak confounding if any from exposure to environmental tobacco

smoke and the other variables listed in Table S2 (online appendix). Radon in the residence is

an additional potential confounder, but information about radon was not available for any

cohort. Radon is likely to be inversely associated with air pollution levels, because radon

concentrations are generally low in apartments, which are common in city areas with higher

air pollution levels. Thus, if confounding by residential radon occurred, we would expect it to

lower the HRs for lung cancer in association with air pollution. Although we adjusted

thoroughly for smoking in all cohorts, we cannot rule out potential residual confounding,

because data on smoking were collected at enrolment, and changes in smoking habits during

18

follow-up were not accounted for. The association was, however, mainly with

adenocarcinoma. If there had been residual confounding, squamous cell carcinomas should

also have been associated with air pollution.

The HRs for lung cancer were similar with and without restriction to participants below most

of the predefined threshold values, indicating that exposure of populations to PM air pollution

even at concentrations below the existing European Union air quality limit values for PM10

(40 µg/m3) and PM2.5 (25 µg/m

3) increases the risk for lung cancer. It is uncertain how widely

the overall risk estimates from this meta-analysis can be generalized to all European

populations, but the absence of significant heterogeneity among the HRs obtained for the

single cohorts indicates that the overall estimate can be generalized.

In conclusion, this very large multicentre study shows an association between exposure to

particulate matter air pollution and the incidence of lung cancer, in particular

adenocarcinoma, in Europe, adding considerably to the weight of the epidemiological

evidence to date.

Contributors

ORN contributed to design, exposure assessment, and interpretation and drafted the

manuscript; ZJA contributed to design, the statistical script and data analyses; RB and KD

contributed to design, exposure assessment, the statistical script and data analyses; ES and

MSt contributed to the statistical script; GW contributed to the statistical script and data

analyses; BH contributed to the statistical script and provided cohort data; PF, MJN, LM,

MK, KTE, TE, ME, KM, MW, KdH, M-YT, AR and CBad contributed to exposure

assessment; BB, KK and PV contributed to design; WWX contributed to design and data

19

analyses; JS, AO, BO, JP, MSø, AV, FR, EM and IT contributed to data analyses; BF, PES,

PN, UDF, NLP, C-GÖ, LF, GP, ATj, PHP, BBdM, TJK, HC, GN, SG, VK, CS, FF, PA, MD

and ATr provided local cohort data; CG and GC contributed to exposure assessment and data

analyses; GH contributed to design, exposure assessment and statistical script. All authors

contributed to critical reading of and comments to the manuscript, interpretation of data and

approved the final draft.

Conflicts of interest

None

20

Reference list

1 Pope III CA, Burnett RT, Thun MJ et al. Lung cancer, cardiopulmonary mortality, and long-term

exposure to fine particulate air pollution. JAMA 2002; 287: 1132-41.

2 Nafstad P, Haheim LL, Oftedal B et al. Lung cancer and air pollution: a 27 year follow up of 16

209 Norwegian men. Thorax 2003; 58: 1071-6.

3 Raaschou-Nielsen O, Andersen ZJ, Hvidberg M et al. Lung cancer incidence and long-term

exposure to air pollution from traffic. Environ Health Perspect 2011; 119: 860-5.

4 Katanoda K, Sobue T, Satoh H et al. An association between long-term exposure to ambient air

pollution and mortality from lung cancer and respiratory diseases in Japan. J Epidemiol 2011; 21:

132-43.

5 Cao J, Yang C, Li J et al. Association between long-term exposure to outdoor air pollution and

mortality in China: a cohort study. J Hazard Mater 2011; 186: 1594-600.

6 Hales S, Blakely T, Woodward A. Air pollution and mortality in New Zealand: cohort study. J

Epidemiol Community Health 2012; 66: 468-73.

7 Lepeule J, Laden F, Dockery D, Schwartz J. Chronic exposure to fine particles and mortality: an

extended follow-up of the Harvard Six Cities study from 1974 to 2009. Environ Health Perspect

2012; 120: 965-70.

8 Heinrich J, Thiering E, Rzehak P et al. Long-term exposure to NO2 and PM10 and all-cause and

cause-specific mortality in a prospective cohort of women. Occup Environ Med 2012;

doi:10.1136/oemed-2012-100876.

9 Yorifuji T, Kashima S, Tsuda T et al. Long-term exposure to traffic-related air pollution and the

risk of death from hemorrhagic stroke and lung cancer in Shizuoka, Japan. Sci Total Environ

2013; 443: 397-402.

10 Cesaroni G, Badaloni C, Gariazzo C et al. Long-Term Exposure to Urban Air Pollution and

Mortality in a Cohort of More than A Million Adults in Rome. Environ Health Perspect 2013.

11 Yorifuji T, Kashima S, Tsuda T et al. Long-term exposure to traffic-related air pollution and

mortality in Shizuoka, Japan. Occup Environ Med 2010; 67: 111-7.

12 Beelen R, Hoek G, van den Brandt PA et al. Long-term exposure to traffic-related air pollution

and lung cancer risk. Epidemiology 2008; 19: 702-10.

13 Raaschou-Nielsen O, Bak H, Sorensen M et al. Air pollution from traffic and risk for lung cancer

in three Danish cohorts. Cancer Epidemiol Biomarkers Prev 2010; 19: 1284-91.

14 Gabrielson E. Worldwide trends in lung cancer pathology. Respirology 2006; 11: 533-8.

15 Chen F, Cole P, Bina WF. Time trend and geographic patterns of lung adenocarcinoma in the

United States, 1973-2002. Cancer Epidemiol Biomarkers Prev 2007; 16: 2724-9.

21

16 Chen F, Jackson H, Bina WF. Lung adenocarcinoma incidence rates and their relation to motor

vehicle density. Cancer Epidemiol Biomarkers Prev 2009; 18: 760-4.

17 Cyrys J, Eeftens M, Heinrich J et al. Variation of NO2 and NOx concentrations between and

within 36 European study areas: Results from the ESCAPE study. Atmos Environ 2012; 62: 374-

90.

18 Eeftens M, Tsai M-Y, Ampe C et al. Spatial variation of PM2.5, PM10 and PM2.5 absorbance and

PM coarse concentrations between and within 20 European study areas and the relationship with

NO2 - results of the ESCAPE project. Atmos Environ 2012; 62: 303-17.

19 Eeftens M, Beelen R, de Hoogh K et al. Development of Land Use Regression models for

PM(2.5), PM(2.5) absorbance, PM(10) and PM(coarse) in 20 European study areas; results of the

ESCAPE project. Environ Sci Technol 2012; 46: 11195-205.

20 Beelen R, Hoek G, Vienneau D et al. Development of NO2 and NOx land use regression models

for estimating air pollution exposure in 36 study areas in Europe - the ESCAPE project. Atmos

Environ 2013; 2013; 72: 10-23.

21 DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986; 7: 177-88.

22 Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med 2002; 21:

1539-58.

23 Beeson WL, Abbey DE, Knutsen SF. Long-term concentrations of ambient air pollutants and

incident lung cancer in California adults: results from the AHSMOG study.Adventist Health Study

on Smog. Environ Health Perspect 1998; 106: 813-23.

24 Turner MC, Krewski D, Pope CA, III et al. Long-term ambient fine particulate matter air pollution

and lung cancer in a large cohort of never-smokers. Am J Respir Crit Care Med 2011; 184: 1374-

81.

25 Vineis P, Hoek G, Krzyzanowski M et al. Air pollution and risk of lung cancer in a prospective

study in Europe. Int J Cancer 2006; 119: 169-74.

26 Benbrahim-Tallaa L, Baan RA, Grosse Y et al. Carcinogenicity of diesel-engine and gasoline-

engine exhausts and some nitroarenes. Lancet Oncol 2012; 13: 663-4.

27 Borm PJ, Schins RP, Albrecht C. Inhaled particles and lung cancer, part B: paradigms and risk

assessment. Int J Cancer 2004; 110: 3-14.

28 Moller P, Jacobsen NR, Folkmann JK et al. Role of oxidative damage in toxicity of particulates.

Free Radic Res 2010; 44: 1-46.

29 Cesaroni G, Porta D, Badaloni C et al. Nitrogen dioxide levels estimated from land use regression

models several years apart and association with mortality in a large cohort study. Environ Health

2012; 11: 48.

30 Nyberg F, Gustavsson P, Jarup L et al. Urban air pollution and lung cancer in Stockholm.

Epidemiology 2000; 11: 487-95.

Table 1. Participants, lung cancer cases, mean air pollution concentrations, and traffic in each cohort.

Cohort

(north-to-

south order)

Nparticipants Mean

age at

baseline

(years)

Nall_LC*

Nadeno†

Nsquam‡

PM10

(µg/m3)

PMcoarse

(µg/m3)

PM2.5

(µg/m3)

PM2.5abs

(10-5

m-1

)

NO2

(µg/m3)

NOx

(µg/m3)

Traffic on

nearest street

(vehicles/day)

Traffic

load on

major

streets

within 100

m (vehicle-

km/day)

EPIC-Umeå,

Sweden 22 136 46·0 69 34 18 NA NA NA NA 5·2 8·7 845 102

HUBRO,

Oslo, Norway 17 640 47·8 75 25 -§ 13·5 4·0 8·9 1·2 20·9 38·3 2502 821

SNACK,

Stockholm,

Sweden 2384 73·1 18 13‖

-

16·4 8·6 8·0 0·8 17·5 33·5 3888 2298

SALT,

Stockholm,

Sweden 4732 57·9 29 12 - 14·9 7·3 7·3 0·6 10·9 18·9 1460 587

Sixty,

Stockholm,

Sweden 3813 60·4 38 22 5 15·0 7·3 7·3 0·6 10·7 18·6 1453 512

SDPP,

Stockholm,

Sweden 7116 47·1 35 22 5 13·6 6·3 6·6 0·5 8·4 14·4 861 110

Table 1

DCH,

Copenhagen,

Denmark 37 447 56·8 638 236 106 17·1 5·7 11·3 1·2 16·3 26·7 2991 1221

EPIC-

MORGEN,

the

Netherlands 15 993 43·7 92 32 24 25·6 8·6 16·9 1·4 23·8 36·5 1535 917

EPIC-

PROSPECT,

the

Netherlands 14 630 57·6 112 43 16 25·3 8·5 16·8 1·4 26·7 39·6 1020 678

EPIC-Oxford,

UK 36 832 45·3 78 19 9 16·1 6·4 9·8 1·1 24·5 40·9 1381 373

VHM&PP,

Voralberg,

Austria 108 018 42·8 678 223 157 20·7 6·7 13·6 1·7 19·9 40·0 1687 294

EPIC-Varese,

Italy 9506 51·6 43 17 12 NA NA NA NA 43·8 86·8 NA NA

EPIC-Turin,

Italy 7216 50·4 48 23 - 46·6 16·6 30·1 3·1 53·0 96·2 3903 465

SIDRIA-

Turin, Italy 4816 44·0 19 - - 48·1 17·0 31·0 3·2 59·8 107·3 4291 810

SIDRIA-

Rome, Italy 9105 44·3 53 - - 36·5 16·7 19·4 2·7 39·1 82·0 2956 1392

EPIC-San

Sebastian,

Spain 7464 49·4 52 - - NA NA NA NA 23·8 47·1 NA 673

EPIC-Athens,

Greece 4096 49·0 18 6 - 45·2 20·8 20·4 2·3 38·0 75·5 9073 11 000

NA, not available

* All lung cancer cases

† Adenocarcinomas of the lung

‡ Squamous cell carcinomas of the lung

§ “-“: No data or too few cases for the model to converge

‖ Contributed to results for adenocarcinomas of the lung among those who lived at the same residence during the whole follow-up period, but did not contribute to the results

for all participants because the model did not converge

Table 2. Associations between six air pollutants and two traffic indicators and the risk for lung cancer; meta-analyses of European

cohorts

HR (95% CI) Measures of heterogeneity

between cohorts

Exposure Increase No. of

cohorts

Model 1* Model 2

† Model 3

‡ Model 3

I2 (%) p-value

PM10 10 µg/m3 14 1·32 (1·12-1·55) 1·21 (1·03-1·43) 1·22 (1·03-1·45) 0·0 0·83

PM2.5 5 µg/m3 14 1·34 (1·09-1·65) 1·17 (0·95-1·45) 1·18 (0·96-1·46) 0·0 0·92

PMcoarse 5 µg/m3 14 1·19 (0·99-1·42) 1·08 (0·89-1·31) 1·09 (0·88-1·33) 33·8 0·11

PM2.5 absorbance 10-5

m-1

14 1·25 (1·05-1·50) 1·09 (0·87-1·37) 1·12 (0·88-1·42) 19·0 0·25

NO2 10 µg/m3 17 1·07 (1·00-1·14) 0·99 (0·93-1·06) 0·99 (0·93-1·06) 0·0 0·70

NOx 20 µg/m3 17 1·08 (1·02-1·14) 1·01 (0·95-1·06) 1·01 (0·95-1·07) 0·0 0·62

Traffic density on

nearest road

5000 vehicles

per day

15 1·02 (0·98-1·06) 1·00 (0·97-1·04) 1·00 (0·97-1·04) 0·0 0·90

Traffic load on major

roads within 100 m

4000

vehicle-km/day

15 1·10 (1·00-1·21) 1·07 (0·97-1·18) 1·09 (0·99-1·21) 0·0 0·92

Table 2

* Model 1: age (time scale in Cox model), sex, calendar time

†Model 2: Model 1 + smoking status, smoking intensity, (smoking intensity)

2, smoking duration, time since quitting smoking,

environmental tobacco smoke, occupation, fruit intake, marital status, educational level, employment status.

‡Model 3: Model 2 + area-level socio-economic status

We included only participants without missing data in any of the variables included in model 3, thus using an identical data set for analyses

with all three models

Table 3. Associations between PM air pollution and the risk for histological subtypes of lung cancer for all participants and for those with the

same residence during the whole follow-up period.

No. of

cohorts

contributing

HR (95% CI)*

HR (95% CI)*

Based on the same cohorts

PM10 PM2.5 PM10 PM2.5

All participants All participants

All lung cancers 14† 1·22 (1·03-1·45) 1·18 (0·96-1·46) All lung cancers 1·22 (1·03-1·45) 1·18 (0·96-1·46)

Adenocarcinomas 11‡

1·49 (1·08-2·05) 1·55 (1·05-2·29) All lung cancers 1·22 (1·01-1·47) 1·16 (0·92-1·45)

Squamous cell carcinomas 7§

0·84 (0·50-1·40) 1·46 (0·43-4·90) All lung cancers 1·19 (0·94-1·51) 1·18 (0·91-1·52)

Table 3

Meta-analysis results based on confounder model 3

* per 10 µg/m

3 PM10 and per 5 µg/m

3 PM2.5

† HUBRO, SNACK, SALT, Sixty, SDPP, DCH, EPIC-MORGEN, EPIC-PROSPECT, EPIC-Oxford, VHM&PP, EPIC-Turin, SIDRIA-Turin, SIDRIA-Rome, EPIC-

Athens

‡ HUBRO, SALT, Sixty, SDPP, DCH, EPIC-MORGEN, EPIC-PROSPECT, EPIC-Oxford, VHM&PP, EPIC-Turin, EPIC-Athens

§ Sixty, SDPP, DCH, EPIC-MORGEN, EPIC-PROSPECT, EPIC-Oxford, VHM&PP

‖ HUBRO, SNACK, SALT, Sixty, SDPP, DCH, VHM&PP, SIDRIA-Turin, SIDRIA-Rome, EPIC-Athens

¶ HUBRO, SNACK, SALT, Sixty, SDPP, DCH, VHM&PP, EPIC-Athens

** Sixty, DCH, VHM&PP

No move during follow-up All participants

All lung cancers 10‖

1·48 (1·16-1·88) 1·33 (0·98-1·80) All lung cancers 1·22 (1·02-1·46) 1·20 (0·96-1·51)

Adenocarcinomas 8¶ 2·27 (1·32-3·91) 1·65 (0·93-2·95) All lung cancers 1·19 (0·98-1·45) 1·17 (0·92-1·49)

Squamous cell carcinomas 3**

0·64 (0·28-1·48) 0·65 (0·16-2·57) All lung cancers 1·21 (0·94-1·55) 1·22 (0·93-1·60)

Table 4. Associations between PM10 and PM2.5 and the risk for lung cancer below thresholds.

Threshold above which

participants were

excluded (µg/m3)

No. of cohorts

contributing to

result

HR (95% CI)* for

the threshold

analyses

HR (95% CI)*

Standard analyses

(no threshold) in the

same cohorts†

PM10 15 5‡ 1·34 (0·51-3·52) 1·21 (0·87-1·68)

20 8§ 1·31 (0·94-1·82) 1·13 (0·92-1·40)

25 10‖

1·17 (0·93-1·47) 1·12 (0·91-1·38)

30 10‖

1·13 (0·92-1·40) 1·12 (0·91-1·38)

35 11¶ 1·11 (0·90-1·37) 1·15 (0·95-1·39)

40 12**

1·13 (0·92-1·39) 1·17 (0·97-1·41)

No threshold 14 (all) ††

1·22 (1·03-1·45) 1·22 (1·03-1·45)

PM2.5 10 6‡‡

1·20 (0·55-2·66) 0·97 (0·63-1·49)

15 8§§

1·11 (0·85-1·45) 1·15 (0·90-1·47)

20 11‖ ‖

1·14 (0·90-1·45) 1·16 (0·92-1·45)

25 11‖ ‖

1·13 (0·90-1·43) 1·16 (0·92-1·45)

No threshold 14 (all) ††

1·18 (0·96-1·46) 1·18 (0·96-1·46)

Meta-analysis results based on confounder model 3

* per 10 µg/m

3 PM10 and per 5 µg/m

3 PM2.5

† Example of reading the table: Ten cohorts contributed to the 30 µg/m

3 threshold analysis for PM10

providing a HR of 1·13. When using the same 10 cohorts for a standard analysis (disregarding thresholds, i.e.

including all participants), the HR was 1·12.

‡ HUBRO, Sixty, SDPP, DCH, EPIC-Oxford

Table 4

§ HUBRO, SNACK, SALT, Sixty, SDPP, DCH, EPIC-Oxford, VHM&PP

‖ HUBRO, SNACK, SALT, Sixty, SDPP, DCH, EPIC-MORGEN, EPIC-PROSPECT, EPIC-Oxford,

VHM&PP

¶ HUBRO, SNACK, SALT, Sixty, SDPP, DCH, EPIC-MORGEN, EPIC-PROSPECT, EPIC-Oxford,

VHM&PP, SIDRIA-Rome

** HUBRO, SNACK, SALT, Sixty, SDPP, DCH, EPIC-MORGEN, EPIC-PROSPECT, EPIC-Oxford,

VHM&PP, EPIC-Turin, SIDRIA-Rome

†† HUBRO, SNACK, SALT, Sixty, SDPP, DCH, EPIC-MORGEN, EPIC-PROSPECT, EPIC-Oxford,

VHM&PP, EPIC-Turin, SIDRIA-Turin, SIDRIA-Rome, EPIC-Athens

‡‡ SNACK, SALT, Sixty, SDPP, DCH, EPIC-Oxford

§§ HUBRO, SNACK, SALT, Sixty, SDPP, DCH, EPIC-Oxford, VHM&PP

‖ ‖ HUBRO, SNACK, SALT, Sixty, SDPP, DCH, EPIC-MORGEN, EPIC-PROSPECT, EPIC-Oxford,

VHM&PP, SIDRIA-Rome

Figure 1. Areas where cohort members lived, measurements were performed, and land-use

regression models for prediction of air pollution were developed

Figure 1

Figure 2. Distribution of particulate matter air pollution at participant addresses in each cohort

Figure 2

Figure 3. Cohort-specific and meta-analysis HRs with 95% CIs for lung cancer incidence in

association with PM10 (per 10 µg/m3) and PM2.5 (per 5 µg/m

3). Based on confounder model 3.

PM10

NOTE: Weights are from random effects analysis

Overall (I-squared = 0.0%, p = 0.828)

EPIC-PROSPECT

EPIC-Oxford

ID

SIDRIA-Turin

SIDRIA-Rome

DCH

EPIC-Turin

Study

EPIC-Athens

SNACK

VHM&PP

Sixty

HUBRO

SALT

EPIC-MORGEN

SDPP

1.22 (1.03, 1.45)

1.89 (0.35, 10.31)

1.64 (0.50, 5.39)

HR (95% CI)

1.41 (0.46, 4.31)

1.35 (0.85, 2.16)

1.10 (0.69, 1.76)

1.45 (0.69, 3.04)

1.55 (1.00, 2.40)

0.89 (0.37, 2.12)

1.20 (0.87, 1.66)

1.63 (0.72, 3.67)

1.06 (0.50, 2.27)

0.69 (0.32, 1.47)

0.36 (0.08, 1.57)

1.17 (0.40, 3.40)

100.00

0.98

1.99

Weight

2.27

12.85

12.77

5.11

%

14.79

3.71

27.70

4.29

4.92

4.82

1.33

2.48

1.22 (1.03, 1.45)

1.89 (0.35, 10.31)

1.64 (0.50, 5.39)

HR (95% CI)

1.41 (0.46, 4.31)

1.35 (0.85, 2.16)

1.10 (0.69, 1.76)

1.45 (0.69, 3.04)

1.55 (1.00, 2.40)

0.89 (0.37, 2.12)

1.20 (0.87, 1.66)

1.63 (0.72, 3.67)

1.06 (0.50, 2.27)

0.69 (0.32, 1.47)

0.36 (0.08, 1.57)

1.17 (0.40, 3.40)

100.00

0.98

1.99

Weight

2.27

12.85

12.77

5.11

%

14.79

3.71

27.70

4.29

4.92

4.82

1.33

2.48

1.25 .5 1 2 4 6

PM2.5

NOTE: Weights are from random effects analysis

Overall (I-squared = 0.0%, p = 0.922)

DCH

SNACK

SIDRIA-Turin

Sixty

EPIC-Athens

EPIC-Turin

SALT

Study

EPIC-MORGEN

EPIC-PROSPECT

SDPP

VHM&PP

SIDRIA-Rome

HUBRO

EPIC-Oxford

ID

1.18 (0.96, 1.46)

0.91 (0.52, 1.60)

0.73 (0.12, 4.37)

1.94 (0.54, 7.00)

1.56 (0.41, 5.98)

0.90 (0.34, 2.40)

1.60 (0.67, 3.81)

1.24 (0.23, 6.76)

0.49 (0.08, 3.21)

1.09 (0.17, 6.99)

2.01 (0.40, 10.01)

1.32 (0.97, 1.81)

1.33 (0.69, 2.58)

0.83 (0.35, 2.00)

0.53 (0.15, 1.91)

HR (95% CI)

100.00

14.09

1.38

2.67

2.45

4.58

5.87

1.54

%

1.26

1.28

1.71

44.56

10.12

5.74

2.73

Weight

1.18 (0.96, 1.46)

0.91 (0.52, 1.60)

0.73 (0.12, 4.37)

1.94 (0.54, 7.00)

1.56 (0.41, 5.98)

0.90 (0.34, 2.40)

1.60 (0.67, 3.81)

1.24 (0.23, 6.76)

0.49 (0.08, 3.21)

1.09 (0.17, 6.99)

2.01 (0.40, 10.01)

1.32 (0.97, 1.81)

1.33 (0.69, 2.58)

0.83 (0.35, 2.00)

0.53 (0.15, 1.91)

HR (95% CI)

100.00

14.09

1.38

2.67

2.45

4.58

5.87

1.54

%

1.26

1.28

1.71

44.56

10.12

5.74

2.73

Weight

1.25 .5 1 2 4 6

Figure 3

Necessary Additional DataClick here to download Necessary Additional Data: Supplementary appendix.pdf